AMD Radeon Pro Vega 56 vs NVIDIA CMP 90HX Comparison

Head-to-Head Benchmarks

The only directly comparable benchmark between the NVIDIA CMP 90HX and the AMD Radeon Pro Vega 56 is Geekbench OpenCL, and the result is decisive. The NVIDIA part scores 69,000 points against AMD’s 61,930 points, a delta of 11.4% in favor of the CMP 90HX. That is a meaningful gap in raw compute throughput, placing the two cards in adjacent but distinct performance tiers.

Context from the nearest rival data reinforces this separation. The CMP 90HX sits at the 90th percentile among all GPUs, with an average benchmark score of 69,000. Its closest competitor, the Intel Arc A770, trails by a razor-thin 0.3% with a score of 68,809. The AMD Radeon Pro WX 8200 and NVIDIA Quadro P6000 actually outpace it slightly — by 1.2% and 1.4%, respectively — meaning the CMP 90HX is not the absolute top of its neighborhood, but it is firmly within striking distance of cards that cost considerably more in their respective lineups.

The Radeon Pro Vega 56, meanwhile, holds an 89th percentile ranking with an average score of 63,693 across three benchmark suites. Its Geekbench OpenCL score of 61,930 is its weakest of the three; it performs better in Geekbench Vulkan at 66,004 and Geekbench Metal at 63,145. This variance matters because it shows the Vega 56’s compute capability is API-dependent, whereas the CMP 90HX’s single recorded score is a flat 69,000 in OpenCL with no other data points to suggest variability.

When comparing the two directly through their nearest rivals, the picture sharpens. The CMP 90HX’s nearest rivals cluster within a narrow band: from 68,562 (AMD Radeon Instinct MI25) up to 69,986 (NVIDIA Quadro P6000). The Vega 56’s nearest rivals cluster lower: from 63,775 (AMD Radeon RX 7600M) down to 64,212 (AMD Radeon Pro WX 9100). The gap between the two clusters is roughly 5,000 points, which is consistent with the 11.4% delta seen in the head-to-head OpenCL test.

Where Each One Wins

The NVIDIA CMP 90HX wins the only test where both cards appear, and it wins by a substantial margin. In Geekbench OpenCL, it delivers 69,000 versus 61,930, an 11.4% advantage. This is the card’s sole benchmark, but it is also the most universal compute test in the database, so the result carries weight. The CMP 90HX also has a higher percentile ranking at 90 compared to the Vega 56’s 89, and its average benchmark score of 69,000 is 8.3% higher than the Vega 56’s 63,693 average.

The AMD Radeon Pro Vega 56, however, shows strengths in other API environments. Its Geekbench Vulkan score of 66,004 is its best result — higher than its OpenCL score by 6.6% and higher than its Metal score by 4.5%. This suggests that in Vulkan workloads, the Vega 56 narrows the gap to the CMP 90HX, though no direct Vulkan comparison exists for the NVIDIA card. The Vega 56’s Metal score of 63,145 is also respectable, though it remains below the CMP 90HX’s OpenCL figure by 9.3%.

For use cases, the data points in different directions. The CMP 90HX is the stronger choice for OpenCL-heavy compute tasks, where its 11.4% lead is unambiguous. The Vega 56, by contrast, offers a more balanced profile across multiple APIs, with its Vulkan performance being particularly notable. If a workload is Vulkan-based, the Vega 56’s 66,004 score suggests it can approach the CMP 90HX’s compute level, though the absence of a direct comparison prevents a definitive ruling.

Architecture Differences

The NVIDIA CMP 90HX is built on the Ampere architecture using the GA102 chip, fabricated on an 8 nm process at Samsung. It packs 28,300 million transistors into a 628 mm² die, yielding a transistor density of 45.1 million per square millimeter. The AMD Radeon Pro Vega 56 uses the older GCN 5.0 architecture with the Vega 10 chip, built on a 14 nm process at GlobalFoundries. It contains 12,500 million transistors on a 495 mm² die, with a density of 25.3 million per square millimeter.

These architectural differences manifest in several ways. The CMP 90HX has 6,400 shading units, 200 texture mapping units, and 80 render output units. It also includes 50 ray tracing cores and 200 tensor cores, features entirely absent from the Vega 56, which has no RT cores and no tensor cores. The Vega 56 counters with 3,584 shading units, 224 TMUs, and 64 ROPs — a lower shader count but a higher TMU-to-shader ratio.

The process node gap is stark: 8 nm versus 14 nm means the Ampere chip achieves nearly double the transistor density (45.1M vs 25.3M per mm²). This density advantage allows the CMP 90HX to fit more than twice the transistors (28,300M vs 12,500M) into a die that is only 27% larger in area. The compute output reflects this: the CMP 90HX delivers 21.89 TFLOPS FP32, whereas the Vega 56 manages 8.960 TFLOPS FP32. For FP16, the CMP 90HX maintains a 1:1 ratio at 21.89 TFLOPS, while the Vega 56 achieves 17.92 TFLOPS via a 2:1 ratio — meaning its FP16 throughput is actually 2.5 times its FP32 rate, but still lower than the CMP 90HX’s raw FP16 figure.

API support also diverges. The CMP 90HX supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The Vega 56 supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3. The difference in DirectX feature level and Vulkan version reflects the generational gap between the two architectures.

Specification Differences

The two cards differ across nearly every specification category. Clock speeds: the CMP 90HX runs at a 1500 MHz base and 1710 MHz boost, while the Vega 56 runs at 1138 MHz base and 1250 MHz boost — the NVIDIA card holds a 29.6% base clock and 36.8% boost clock advantage. Memory: the CMP 90HX has 10 GB of GDDR6X on a 320-bit bus with 760.3 GB/s bandwidth and a 1188 MHz memory clock (19 Gbps effective). The Vega 56 has 8 GB of HBM2 on a 2048-bit bus with 402.4 GB/s bandwidth and a 786 MHz memory clock (1572 Mbps effective). The CMP 90HX’s bandwidth advantage is 89% higher.

Pixel and texture rates follow the same pattern. The CMP 90HX achieves 136.8 GPixel/s and 342.0 GTexel/s, versus 80.00 GPixel/s and 280.0 GTexel/s for the Vega 56 — leads of 71% and 22%, respectively. Power draw differs significantly: the CMP 90HX is rated at 320 W TDP with dual-slot cooling and requires two 8-pin power connectors plus a 700 W suggested PSU, while the Vega 56 is an integrated GPU (IGP) with no power connectors and a 210 W TDP.

The bus interfaces are notably different. The CMP 90HX uses PCIe 1.0 x4, which is an unusual choice that limits host transfer bandwidth, while the Vega 56 uses PCIe 3.0 x16. Display outputs: the CMP 90HX has none, reflecting its mining-oriented design, whereas the Vega 56 provides 1x HDMI 2.0b and 3x DisplayPort 1.4a. Physical dimensions: the CMP 90HX measures 285 mm by 112 mm, while the Vega 56’s dimensions are not listed. Release dates are separated by nearly four years: the CMP 90HX launched on 2021-07-27, the Vega 56 on 2017-08-13. Both are end-of-life.

FAQ

Q: Which card has a higher Geekbench OpenCL score?

A: The NVIDIA CMP 90HX scores 69,000, which is 11.4% higher than the AMD Radeon Pro Vega 56’s 61,930.

Q: Does the AMD card perform better in any benchmark?

A: The Vega 56 has no direct benchmark win over the CMP 90HX. However, its Geekbench Vulkan score of 66,004 is its best result and exceeds its own OpenCL score by 6.6%.

Q: What are the memory configurations of each card?

A: The CMP 90HX has 10 GB of GDDR6X on a 320-bit bus with 760.3 GB/s bandwidth. The Vega 56 has 8 GB of HBM2 on a 2048-bit bus with 402.4 GB/s bandwidth.

Q: Which card has more shading units?

A: The NVIDIA CMP 90HX has 6,400 shading units, compared to 3,584 on the AMD Radeon Pro Vega 56.

Q: Do either of these cards support ray tracing?

A: Only the CMP 90HX includes ray tracing cores (50) and tensor cores (200). The Vega 56 has neither.

Q: What is the power consumption difference?

A: The CMP 90HX is rated at 320 W TDP, while the Vega 56 is rated at 210 W TDP.

The Verdict

The data points to a clear performance hierarchy. The NVIDIA CMP 90HX outscores the AMD Radeon Pro Vega 56 by 11.4% in their only shared benchmark, holds a higher percentile ranking (90 vs 89), and delivers more than double the FP32 compute (21.89 TFLOPS vs 8.960 TFLOPS). Its memory bandwidth advantage is nearly 89%, and it offers features the Vega 56 lacks entirely, including ray tracing and tensor cores. The CMP 90HX also has a higher boost clock (1710 MHz vs 1250 MHz) and more shading units (6,400 vs 3,584).

The AMD Radeon Pro Vega 56, however, has its own merits. It is far more power-efficient per the listed TDP, consuming 210 W versus 320 W. It is an integrated GPU with no power connectors, making it suitable for systems where discrete power cabling is unavailable. It provides display outputs, unlike the CMP 90HX which has none. Its Vulkan performance (66,004) suggests it can compete in API-specific workloads, and its HBM2 memory with a 2048-bit bus offers a different memory architecture that may suit certain applications.

Who should pick which? The CMP 90HX is the choice for compute-heavy workloads where OpenCL performance is the primary metric and power draw is not a constraint. The Vega 56 is the choice for systems requiring integrated graphics, display output, and lower power draw, particularly if Vulkan workloads are the main use case. The CMP 90HX wins on raw performance; the Vega 56 wins on flexibility and efficiency. Neither card is currently in production, but for those choosing between the two based on benchmark data alone, the NVIDIA part is the stronger performer.